Please read this first. Voice analysis is a research-stage wellbeing signal, not a diagnosis and not a medical or psychiatric test. It does not detect, diagnose, screen for, identify, or predict depression, anxiety, suicidal thoughts, or any mental-health condition, and no consumer app, Sonora included, can do so. This page is general information, not medical advice. If you are worried about your mental health, please speak to a GP or a qualified mental-health professional. If you are in distress or crisis, the support is real and there is a clear signpost near the end of this page.
Can your voice reveal your mental state?
The honest answer has to start with what voice analysis cannot do, because that is the part most likely to worry someone who arrives here anxious about themselves. Voice analysis cannot tell you whether you have depression, anxiety, or any other condition. It is not a diagnosis, not a screening test, and not a medical assessment of your mind. What the research actually shows is narrower. In studies, certain measurable features of how people sound, such as the pitch and steadiness of the voice, tend to shift on average when groups of people are stressed or low in mood. That is a correlation seen across many people, at a population level, in research conditions. It is a weak, probabilistic signal, and it is still early-stage science. It cannot be turned around to tell one individual what is happening inside them.
It helps to hold two ideas at once. Your voice does carry information about how you feel, in the everyday sense that a friend can often hear when you sound tired or tense. And, separately, no software can read that everyday signal precisely enough, or reliably enough, to make a clinical judgement about your mental health. This page walks through what researchers have genuinely found, why a research signal is not the same as a personal verdict, how Sonora does and does not use voice, and what to do if you are worried. The thread running through all of it is simple: this is information to understand a technology, never a tool to assess yourself.
For the wider context of how voice reading fits into personalised, adaptive audio, and the evidence behind that whole approach, read the AI sound therapy pillar, which sets out the full picture and its honest limits.
What the research actually links to voice
Researchers studying speech have found that some acoustic features of the voice, the measurable qualities of the sound rather than the words, move with stress and mood. A 2025 systematic review of acoustic correlates in speech concluded that certain vocal features do track with negative emotion and stress, while being candid that these signals vary a great deal between people and settings, and that people can mask how they feel, which makes any single reading unreliable on its own.1 That candour is the point: the link is real at the level of averages across many people, and it is noisy at the level of any one person.
Some of this work connects voice to the body's own stress response. A 2025 study compared speech features against salivary cortisol, the stress hormone, in volunteers put through a controlled stressor, and found that some vocal measures tracked the genuine physiological response, supporting the idea of voice as a non-invasive stress signal.2 Even here the effects varied between people. It is just as important to report where the evidence is shakier. A 2025 systematic review and meta-analysis of voice pitch as a stress marker found that pitch does tend to rise after stress, but cautioned that once the analysis was corrected for publication bias the effect was no longer statistically reliable, and called for validation in large, prospective studies before voice pitch is treated as a standalone biomarker.3
So what does research link to voice, fairly stated? Everyday stress and low mood, at a group level, through weak and variable signals, in research settings, using tools that are not yet validated for clinical use. These are sub-clinical signals: indicators of ordinary states like tension or tiredness, not measurements of any disorder. The deep dive into the specific voice features is a separate guide; see Vocal Biomarkers Explained (publishes when that cluster ships).
Screening versus diagnosis: a crucial difference
This is the most important section on the page, so it is worth slowing down. Two words get used loosely in this area, and the difference between them matters enormously. Screening means a rough first pass that flags whether someone might benefit from a closer look; it is never a conclusion. Diagnosis is the considered judgement a qualified clinician reaches after assessment, using their training, a proper conversation, your history, and clinical guidelines. Voice analysis, at its absolute best and only in research, sits at the very soft end of the first category. It is, at most, a faint wellbeing signal. It is nowhere near the second. Nothing about reading a voice produces a diagnosis, and no honest description of the technology would claim it does.
The phrase you will sometimes see for this research area is digital phenotyping: using everyday digital signals, including speech, to gauge something about a person's state over time. It is an active research field, but the same review evidence applies. Speech-based approaches to mental health are described by researchers as promising yet still developing, tested mostly on small samples, and far from routine clinical use.4 Researchers have explored, for example, whether speech features might one day help track depression: an early and much-cited study found that speech in depression showed slower pace and longer pauses, and that these eased as people responded to treatment.5 That is genuine, careful work, and it still does not put a diagnostic tool in anyone's pocket. The right takeaway is firm and reassuring at once: only a qualified professional can assess and diagnose mental health, and that is exactly as it should be. A voice signal is, at most, a prompt to have a proper conversation with a person trained to have it.
Digital biomarkers and where this is heading
A digital biomarker is a measurable signal collected through everyday technology, such as a phone or a voice recording, that researchers hope might tell us something useful about health or state. Voice is one candidate among many. The promise is real: a non-invasive signal that needs no clinic visit and no blood test is genuinely appealing, and the research is moving quickly. But the honest account sits alongside the promise. The signals are weak and noisy, they vary between individuals, and a marker that looks encouraging in one study can fade once it is tested more rigorously, as the voice-pitch meta-analysis above showed when the effect did not survive correction for publication bias.3
There are deeper concerns too. Models trained mostly on one group of people can perform worse for others, so bias and fairness are live issues for any voice-based tool. Reliability is unsettled, since how you sound depends on a cold, a bad night, an accent, the room, or the microphone, none of which has anything to do with your mental health. And privacy matters whenever a device listens to you. The field's own reviewers consistently call for larger, longer studies before any of this approaches clinical use.4 The fair summary is that voice-based digital biomarkers are an interesting research direction with serious unsolved problems, not a shortcut to understanding your own mind, and not a substitute for a professional.
How Sonora uses voice signals (and how it doesn't)
Given all of that, it is worth being precise about what Sonora actually does. Sonora is a wellbeing app that personalises calming, focusing, or sleep-supporting audio. It listens to a short voice sample to pick up everyday markers of how you sound, such as stress, fatigue, and energy, and then shapes a soundscape toward the state you have asked for. That voice read has one purpose only: to match the audio to your everyday mood, in the same spirit as choosing gentler music when you sound tired.
Here is the firm boundary. Sonora does not, and is not designed to, detect, diagnose, screen for, identify, or predict any mental-health condition. It is not a medical device, a psychiatric test, or a clinical tool of any kind, and it produces no assessment of your mental health for you to read or worry about. The voice signal it uses is the everyday, sub-clinical kind described throughout this page, not a clinical measurement. If anything you are feeling concerns you, Sonora is not the place to look for answers about it; a qualified professional is. The app is a wellbeing aid, and it is honest about being only that. You can see the full evidence base behind Sonora's wider claims on Sonora's published evidence base.
When to seek professional support
If you are worried about your mental health, or you have been feeling low, anxious, or not yourself for a while, please talk to someone who can properly help. The clearest first step is to speak to a GP or a qualified mental-health professional, who can listen, assess, and point you to the right support; the National Health Service sets out how to find that help.6 In the United States, the National Institute of Mental Health similarly advises starting with a health-care provider, who can carry out an initial mental-health screening and refer you to a professional.7 Reaching out is a strong, sensible thing to do, and these services exist precisely so you do not have to work things out alone or from an app.
If you are in distress right now, or having thoughts of harming yourself, please get help straight away rather than waiting. In the United Kingdom you can call the Samaritans free, at any time, on 116 123; if you have seriously harmed yourself or feel you may be about to, the NHS advises calling 999 or going straight to A&E, and you can call NHS 111 to find urgent help.8 Wherever you are, you can contact your GP, your local emergency services, or a recognised crisis line. You deserve real support from a person, and it is available.
Related guides
This is one of several AI sound therapy guides in the Sonora Learn library. For how a voice recording is turned into a signal, step by step, read How AI Voice Analysis Works. For the specific acoustic features behind that process, see Vocal Biomarkers Explained (publishes when that cluster ships). For the broader approach and its evidence, read the AI sound therapy pillar.
You can find all articles in the Learn library. If you take one thing from this page, let it be this: voice analysis is an early-stage research signal, not a verdict on your mind, and real help from a professional is always the right call when you are worried.
